Open zjuchenyuan opened 5 years ago
only 4 of 30 finished successfully.
Why did they fail?
Do I need to follow this warning to change the code? If I do, does this change impact fuzzing performance when fuzzing other programs?
My suggestion: If the density is > 50%, try to disable context. If the density is 10%~50%, try to modify MAP_SIZE.
Considering there are tremendous queue files, you 'd better disabling context.
In short, when fuzzing pdftotext, Angora generates tremendous amount of queue files, making analyze queue coverage infeasible.
Here is the detailed experiment setup:
The
pdftotext
is compiled using xpdf-4.00. Parallel running 30 independent instance, each bind to a certain CPU using cpuset, limit memory to 2GB, and want to fuzz 24 hours. The angora image is based on your Dockerfile, but change the config fileAngora/common/src/config.rs
to setMAX_INPUT_LEN
to 1MB. (so I call itangora_no15k
)But most of the instances exited before 24 hours due to memory limit (being killed), running time are given below, only 4 of 30 finished successfully.
You can see the queue files are too many, which makes it hard to analyze coverage using afl-cov.
Besides, density too large warning:
Do I need to follow this warning to change the code? If I do, does this change impact fuzzing performance when fuzzing other programs?
Any suggestions? How to correctly using Angora to fuzz pdftotext? (I know I need to rerun this fuzzing experiment without memory limit